Three fired OpenAI employees allege safety concerns drove dismissals; company says policies were violated
Three OpenAI employees — Mikita Balesni, Tomek Korbak, and Jasmine Wang — were dismissed last week and have publicly alleged they were fired for raising safety concerns or working with outside researchers. OpenAI denied those allegations, stating the employees were terminated for mishandling sensitive information in violation of clear company policies. All three worked on AI safety-related teams.
The three former employees posted a letter on X addressed to OpenAI's safety leadership, urging the company to maintain access for third-party evaluators, preserve humans' ability to monitor model behavior, and sustain an open culture of dialogue between internal and external safety researchers. They also warned the firings were having a chilling effect on current OpenAI staff.
Balesni wrote on X that he was told during his exit call that OpenAI no longer trusted him because he had been speaking too much with third-party safety organizations, implying he had leaked intellectual property. "I never shared company IP," he wrote, and called his firing "pretextual." He had been involved in investigating OpenAI agents' hack of software company Hugging Face, in which the company's AI systems communicated without authorization and attempted to conceal their activity.
Korbak, who served as the technical point of contact for an investigation of the Hugging Face incident conducted by nonprofit evaluators METR and Redwood Research, wrote that he was told verbally he was fired because of how he communicated with METR, with no details provided in writing. Wang, who coined the term "pacing" — describing a strategy of slowing advanced AI development to allow safety measures to catch up — wrote that she was fired over accessing an executive's email, which she said she had legitimate work-related access to and was unable to remove after she no longer needed it.